ANI Models
The ANI (Accurate NeurAl networK engINe for Molecular Energies) family is a set of neural network potentials developed by the Roitberg group. ANI models are among the fastest ML potentials available, making them ideal for high-throughput screening and rapid prototyping of neutral organic molecular systems.
Available Variants
| Model | Keyword | Elements | Training Level | Best For |
|---|---|---|---|---|
| ANI-2x | ani2x |
H, C, N, O, F, S, Cl | ωB97X / 6-31G(d) | General organic chemistry with halogens |
| ANI-1x | ani1x |
H, C, N, O | ωB97X / 6-31G(d) | Fast calculations on HCNO-only systems |
| ANI-1ccx | ani1ccx |
H, C, N, O | CCSD(T)/CBS (transfer-learned) | Publication-quality accuracy for small molecules |
| ANI-1xnr | ani1xnr |
H, C, N, O | ωB97X / 6-31G(d) (non-reactive) | Equilibrium geometries without bond-breaking |
DFT-D4 Dispersion Correction
ANI models are the only models in MAPLE that support the DFT-D4 dispersion correction. D4 adds London dispersion (van der Waals) interactions, which is critical for non-covalent binding, molecular stacking, and conformational energetics.
#model=ani2x
#d4
#opt(method=lbfgs)
Tip
Always enable #d4 when studying systems with significant non-covalent interactions (hydrogen bonds, pi-stacking, host-guest complexes). The computational overhead is minimal.
Calculator Capabilities
- Hessian: analytic and numerical Hessians are available in gas phase. Implicit-solvent Hessians (analytic or numerical) are not supported.
- HVP: ANI is the only shipped backend that implements
get_hvp()for the HVP-enabled Dimer path. ANI HVP is gas-phase only. - Implicit solvent: MAPLE's experimental GB-polar/QEq correction is energy-only (
#spwithexperimental=true); solvent forces, HVP, and Hessians are not supported. - Model files: MAPLE can auto-download ANI checkpoints from the MAPLE model repository, or you can provide an explicit
model_path.
Usage Examples
General organic optimization
#model=ani2x
#device=gpu0
#opt(method=lbfgs)
0 1
C 0.000000 0.000000 0.000000
O 1.200000 0.000000 0.000000
H -0.540000 0.940000 0.000000
H -0.540000 -0.940000 0.000000
High-accuracy with coupled-cluster potential
#model=ani1ccx
#device=gpu0
#opt(method=lbfgs, level=extratight)
0 1
O 0.000000 0.000000 0.117300
H 0.000000 0.756900 -0.469200
H 0.000000 -0.756900 -0.469200
With D4 dispersion for non-covalent interactions
#model=ani2x
#d4
#device=gpu0
#opt(method=lbfgs, level=tight)
XYZ /path/to/dimer_complex.xyz
Limitations
- No charge/multiplicity support. ANI models assume neutral, singlet ground state; use AIMNet2 or UMA
omolfor charged/open-shell inputs. - Limited elements. ANI-2x covers 7 elements; ANI-1x/1ccx/1xnr cover only H, C, N, O.
- No periodic systems. ANI models do not support periodic boundary conditions.
- Implicit solvation is energy-only. MAPLE's experimental GB-polar/QEq correction can be requested for
#spwithexperimental=true; solvent forces, optimization, MD, TS, scan, and frequency workflows remain unsupported for that correction.
